Module completion checklist

Data prep for ggplot2

Creating complex plots using ggplot2

• We have already created a simple histogram and scatterplot for a single variable from our dataset

• But the next step is to compare the distributions of variables

• To do so, we will create a series of normalized boxplots, allowing us to compare each variable’s relative variation in data

• We will then create a complex scatterplot of the normalized values

Set up: load tidyverse

# Load tidyverse library
library(tidyverse)

# Save our custom `ggplot` theme to a variable.
my_ggtheme = theme_bw() +                     
  theme(axis.title = element_text(size = 20),
        axis.text = element_text(size = 16),
        legend.text = element_text(size = 16),              
        legend.title = element_text(size = 18),             
        plot.title = element_text(size = 25),               
        plot.subtitle = element_text(size = 18))          

Set up: data conversion with gather

health_subset_long = health_subset %>%
  gather(key = "variable",
         value = "value") 

# Inspect the first few observations.
head(health_subset_long)

  variable value
1      age    67
2      age    61
3      age    80
4      age    49
5      age    79
6      age    81

# Inspect the last few observations.
tail(health_subset_long)

      variable    value
15325      bmi 18.60000
15326      bmi 28.89324
15327      bmi 40.00000
15328      bmi 30.60000
15329      bmi 25.60000
15330      bmi 26.20000

Set up: tidy data look

# Inspect the first few entries in the data.
head(health_subset_long)

  variable value
1      age    67
2      age    61
3      age    80
4      age    49
5      age    79
6      age    81

# Inspect the last few entries in the data.
tail(health_subset_long)

      variable    value
15325      bmi 18.60000
15326      bmi 28.89324
15327      bmi 40.00000
15328      bmi 30.60000
15329      bmi 25.60000
15330      bmi 26.20000

Module completion checklist

Set up & link data: make boxplots

• To add a boxplot layer to the base plot, use geom_boxplot()

# A basic box plot with pre-saved theme.
boxplots = ggplot(health_subset_long,   #<- set the base plot + data
                  aes(x = variable,  #<- map `variable` to x-axis
                      y = value)) +  #<- map `value` to y-axis
           geom_boxplot() +          #<- add boxplot geom
           my_ggtheme                #<- add pre-saved theme

Set up & link data: make boxplots (cont’d)

# View plot.
boxplots

Set up: normalize data with group_by + mutate

• The variables are on different scales, making it harder to compare them
• If we normalize the variables, we can more easily assess their relative distributions with respect to a numerical standard

# Let's normalize the data and then create boxplots.
health_subset_long = health_subset_long %>%
  group_by(variable) %>%              #<- group values by variable
  mutate(norm_value =                 #<- make `norm_value` column
           value/max(value,           #<- divide value by group max
                     na.rm = TRUE))   #<- don't forget the NAs!

head(health_subset_long)

# A tibble: 6 x 3
# Groups:   variable [1]
  variable value norm_value
  <chr>    <dbl>      <dbl>
1 age         67      0.817
2 age         61      0.744
3 age         80      0.976
4 age         49      0.598
5 age         79      0.963
6 age         81      0.988

Set up: boxplot with normalized data

• Let’s save the boxplot with normalized data to a variable
• Add my_ggtheme to it

# Construct the base plot for normalized data.
base_norm_plot = ggplot(health_subset_long,        #<- set data
                        aes(x = variable,       #<- map `variable`
                            y = norm_value)) +  #<- map `norm_value`
                 geom_boxplot() +               #<- add boxplot layer
                 my_ggtheme                     #<- add theme

Set up: boxplot with normalized data (cont’d)

# View 
base_norm_plot

Adjust: normalized boxplot aesthetics

• Working in black and white alone, comparing the distributions of the variables may be more challenging

# Let's add a fill color to the plot.
boxplots_norm = ggplot(health_subset_long,        #<- set data
                        aes(x = variable,       #<- map `variable`
                            y = norm_value,     #<- map `norm_value`
                            fill = variable)) + #<- set fill
                 geom_boxplot()  +              #<- boxplot layer
                 my_ggtheme                     #<- add theme

Adjust: normalized boxplot aesthetics (cont’d)

# View boxplot
boxplots_norm

Adjust: normalized boxplot legends

• To remove the legend for fill color, use the function guides() and pass fill = FALSE to it
• Click here to read more about modifying legends with guides()

# Make color of fill based on variable.
boxplots_norm = boxplots_norm +     #<- set base plot   
                guides(fill = FALSE) #<- remove redundant legend 

Adjust: normalized boxplot legends (cont’d)

# View updated plot.
boxplots_norm

Polish: normalized boxplot details

• Finally, we can polish the visualization by adding features for improved readability and usability

# Make outliers stand out with red color and bigger size.
boxplots_norm = boxplots_norm +       #<- previously saved plot
  geom_boxplot(outlier.color = "red", #<- adjust outlier color
               outlier.size = 5) +    #<- adjust outlier size
  labs(title = "Health data variables",  #<- add title and subtitle
       subtitle = "Boxplot of scaled data")

Polish: normalized boxplot details (cont’d)

# View updated plot.
boxplots_norm

• Chat question: What insights can you gather based on this visualization?

Building a scatterplot with ggplot2

• Now that we have boxplots of the normalized variables, we can start to visualize their relationship
• We will use scatterplots to demonstrate such relationships, but we must prepare and transform our data again before building them
  • Separate age from all other variables
  • Ensure that each entry in age corresponds to an entry in all other variables
  • Convert the original wide data to a long format excluding the age variable

Set up: transform data for scatterplot

health_subset_long2 = health_subset %>%
  gather(avg_glucose_level:bmi, #<- gather all variables but `age`
         key = "variable",                            #<- set key to `variable`
         value = "value") %>%                         #<- set value to `value`
  group_by(variable) %>%
  mutate(norm_value = value/max(value, na.rm = TRUE))

# Inspect the data.
head(health_subset_long2)

# A tibble: 6 x 4
# Groups:   variable [1]
    age variable          value norm_value
  <dbl> <chr>             <dbl>      <dbl>
1    67 avg_glucose_level  229.      0.842
2    61 avg_glucose_level  202.      0.744
3    80 avg_glucose_level  106.      0.390
4    49 avg_glucose_level  171.      0.630
5    79 avg_glucose_level  174.      0.641
6    81 avg_glucose_level  186.      0.685

Set up & link: normalized data base plot

# Create a base plot.
base_norm_plot = ggplot(data = health_subset_long2,  #<- set data 
                        aes(x = norm_value,       #<- set x-axis to represent normalized value
                            y = age,            #<-   y-axis to represent `age` 
                            color = variable)) +  #<- set color to depend on `variable`
                        my_ggtheme                #<- set theme

base_norm_plot

Set up & adjust: normalized data scatterplot

# Create a scatterplot.
scatter_norm = base_norm_plot +        #<- base plot
               geom_point(size = 3,    #<- add point geom with size of point = 3
                          alpha = 0.7) #<-  make it 70% opaque

# View updated plot.
scatter_norm

Adjust: add density geom to scatterplot

• Now add a 2D density geom to our normalized scatterplot

# Adjust scatterplot to include 2D density.
scatter_norm = scatter_norm +              #<- previously saved plot
               geom_density2d(alpha = 0.7) #<- add 2D density geom with 70% opaque color

# View updated plot.
scatter_norm

Adjust: wrap scatterplots in facets

• Notice that the previous plot is hard to decipher because of the overlapping data
• Instead of overlaying densities, let’s split them into individual plots called facets
• These facets will appear as a tiled matrix of related visualizations
• We create them using the facet_wrap() function, which splits the data by one or more variables and plots the subsets together
• Click here to learn more about facets on the tidyverse website

# Wrap each scatterplot into a facet.
scatter_norm = scatter_norm +                   #<- previously saved plot
               facet_wrap(~ variable, ncol = 3) #<- wrap plots by variable into 3 columns

Adjust: wrap scatterplots in facets (cont’d)

# View updated plot.
scatter_norm

Adjust & polish: legends and text in scatterplot

# Add finishing touches to the plot.
scatter_norm = scatter_norm +                        #<- previously saved plot
  guides(color = FALSE) +                            #<- remove legend for color mappings
  theme(strip.text.x = element_text(size = 14)) +    #<- increase text size in strips of facets
  labs(title = "Health data: Age vs. other variables",#<- add title and subtitle
       subtitle = "2D distribution of scaled data")

Adjust & polish: legends and text in scatterplot

# View updated plot.
scatter_norm

Saving plots

• Once we have created a plot in R, we may want to save it to a file so we can use it in another document
• The first step in saving plots is to decide which output format we want to use
• The available file formats are PNG, JPEG, BMP, and PDF
  • We will explore the PNG and PDF formats

Saving plots: export to PNG

png("Name_of_file.png", #<- name of file
    width = 400,        #<- width of image
    height = 300,       #<- height of image
    units = "px") #<- units for height & width

plot 1   #<- call the plot object you want to export

dev.off() #<- closes R graphics device

Saving plots: export to PNG (cont’d)

# Set working directory 
# to where we want to save our plots.
setwd(plot_dir)

png("Heart_boxplots_norm.png",
    width = 1200, 
    height = 600, 
    units = "px")
boxplots_norm
dev.off()

RStudioGD 
        2 

setwd(plot_dir)
png("Heart_scatterplot_norm.png",
    width = 1200, 
    height = 600, 
    units = "px")
scatter_norm
dev.off()

RStudioGD 
        2 

Saving plots: export to PDF

• There are a few advantages of saving plots to a PDF format as opposed to an image:

  • PDF documents allow R’s native vector graphics to shine: the image quality will not be affected if we zoom in
  • We can save multiple plots into a single PDF document with one command

Saving plots: export to PDF (cont’d)

pdf("Name_of_file.pdf", #<- name of file
    width = 16,         #<- width in inches
    height = 8)         #<- height in inches

plot 1    #<- plot object or plotting function
plot 2    #<- plot object or plotting function
...       #<- arbitrary number of plots

dev.off() #<- clear graphics device

Saving plots: export to PDF (cont’d)

• Let’s save more than one plot together in a PDF

# Set working directory to where you want to save plots.
setwd(plot_dir)

pdf("health_plots.pdf", #<- name of file
    width = 16,      #<- width in inches
    height = 8)      #<- height ...

boxplots_norm        #<- plot 1
scatter_norm         #<- plot 2

dev.off()            #<- clear graphics device

RStudioGD 
        2 

Knowledge Check

Exercise

Module completion checklist

Static Plots: topic summary

In this part of the course, we have covered the following concepts:

• Discuss the process of constructing complex plots
• Visualize data with ggplot2

Congratulations on completing this module!